Is is possible to use RankLib with more than 5 relevance labels in my data? I have measurements that form queries for 96 examples at a time, but the scores (relevance labels) are not discrete -- less so in 5 categories.
This is something that I have been able to do using SVMrank (http://www.cs.cornell.edu/people/tj/svm_light/svm_rank.html).
Thanks for your help in advance.
Thanks,
Shyam
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes you can use more of fewer labels if you want. Five is the default (values 0-4).
Do you mean you are using a score for a relevance label (non-integer value)?
The max number of relevance labels (-gmax argument) gets converted to an integer, but is actually used as a decimal normalizing value for the evaluation metric, so I'm not really certain of the effects of using non-discrete labels.
It might have an effect on evaluation metric characteristics, but it won't blow up!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
Is is possible to use RankLib with more than 5 relevance labels in my data? I have measurements that form queries for 96 examples at a time, but the scores (relevance labels) are not discrete -- less so in 5 categories.
This is something that I have been able to do using SVMrank (http://www.cs.cornell.edu/people/tj/svm_light/svm_rank.html).
Thanks for your help in advance.
Thanks,
Shyam
Yes you can use more of fewer labels if you want. Five is the default (values 0-4).
Do you mean you are using a score for a relevance label (non-integer value)?
The max number of relevance labels (-gmax argument) gets converted to an integer, but is actually used as a decimal normalizing value for the evaluation metric, so I'm not really certain of the effects of using non-discrete labels.
It might have an effect on evaluation metric characteristics, but it won't blow up!
Thanks for your reply.
Right, exactly. I am using a score for the relevance label.
I guess I will try it and then post again with the outcome in case it could be useful to others in the future.
Thanks!
Shyam